Conquering the Algorithms and Developing the First Generation Artificial PancreasFriday, May 04, 2012
By borrowing a method used in engineering to help develop the complex algorithms needed for varying human insulin requirements, Biomathematician Boris Kovatchev, PhD, has played an instrumental role in helping to bring about the first-generation artificial pancreas. The University of Virginia (UVA) researcher and his collaborators just recently started the first U.S. outpatient trial testing his prototype.
By: John Parkinson, Clinical Content Coordinator, DiabetesCare.net
One of the more exciting developments in type 1 diabetes research in recent years has been the scientific work surrounding the artificial pancreas.
This idea of taking three components, a continuous glucose monitor (CGM), insulin pump, and a device that talks to the CGM and the pump and gives the go-ahead to deliver insulin in a closed-loop system has been around since the mid-2000s. The stopping point to creating a closed loop system, however, was developing the real-time algorithms to have the artificial pancreas deliver insulin at the right time and in the correct dosage.
That challenge may be over.
Dr. Kovatchev (pictured here) collaborated with UVA colleagues and fellow researchers at the University of Padova in Italy to create the necessary algorithms—in fact they created three altogether—to allow for the artificial pancreas to regulate a person’s insulin levels. The most advanced algorithm is set up to automatically deliver the correct amount of insulin to the person without any manual instructions, thus creating a truly closed loop device.
Back in the mid-2000s, there were some who were skeptical the algorithms could be written for people in real time, understanding there was a time-delay in the existing CGM technology and glucose readings. Contrary to this common belief, Dr. Kovatchev surmised it could be done and went to work to overcome this monumental biomathematical challenge.
If anyone was in the right position to do this research, Dr. Kovatchev certainly has the pedigree. His expertise lies in biomathematics, specifically modeling biologic and behavioral processes—certainly apt for tackling the algorithms associated with the artificial pancreas.
He has also been working in diabetes translational research for years, developing 36 U.S. and international patents, and Dr. Kovatchev has 62 patents still pending. He has won both the Diabetes Technology Leadership Award in 2008 and the Edich-Henderson Inventor of the Year Award in 2011.
Lastly, he, like many in the diabetes industry, has a personal connection to it: his father suffered from diabetes for countless years.
Dr. Kovatchev was in part able to achieve his accomplishment by taking a different approach in carrying out clinical trials. His team at UVA, along with Dr. Claudio Cobelli’s team at the University of Padova, created a human metabolic simulator. By doing this, they were able to then test medical algorithms in silico–with simulated “people” with type 1 diabetes in place of doing animal trials.
While using simulators is quite unusual in medical research, this approach is frequently used in engineering. And Dr. Kovatchev believes using a human simulator decreased the amount of time by years it would have taken to get to their present outpatient human trials.
DiabetesCare.net recently sat down with Dr. Kovatchev to discuss his algorithm work, how the artificial pancreas works, the remaining challenges to it, and when an artificial pancreas may be available in the marketplace.
DiabetesCare.net: What specific technological components does this artificial pancreas encompass?
Kovatchev: There are three principal components: one is the CGM and it can be subcutaneous or implantable; an insulin pump; and the smart phone, which is the tool that bridges the two. It reads the sensor, calculates the insulin, and tells the pump to deliver it.
(The smart phone component of the artificial pancreas)
DiabetesCare.net: Do you believe you have created the first-generation artificial pancreas?
Kovatchev: I think we have created the first-generation portable artificial pancreas. It has been in existence for seven months now and it has been tested in 400 plus hours of use. It is based on algorithms that have been tested over years.We’re encouraged that the cell phone can be the computational platform for the artificial pancreas, and we just need to work on the wireless connectivity a bit more.
The connectivity is still a work in progress.The current device does not have a wireless connection. Since we don’t have this, we got an experimental device from a company, Insulet, and the device has a short-range wireless connection to a Dexcom sensor and the insulin pump. We used this device and attached a small box with a wire to it and it transmits a signal to the cell phone.
The pump and cell phone are items that were off the shelf. There is a little bit of work that has to be done, but we hope to remove the box very shortly. We are looking at having that be omitted from the system within a couple of months.
This is just an engineering issue and it won’t require any specific human clinical trials to overcome. Any device can have a Bluetooth chip.
DiabetesCare.net: When you say the wire and the box, do you mean that it is connected to the smart phone?
Kovatchev: No, the smart phone is completely wireless; the wire and the box covert the signal from the sensor and the pump into bluetooth. The wire and the box have to be close to the person because of the needs of the sensor. The person wears the sensor and the pump. We placed a little pouch on the patient and he carried the box and the wire in the pouch. The cell phone does not have to be on the person but nearby.
DiabetesCare.net: Can you provide an overview of how your artificial pancreas will work for people using them?
Kovatchev: The entire algorithm system will be implemented in the cell phone and having a wireless connection to the pump and the CGM. We have taken a normal cell phone off the shelf, modified it, and the phone talks to the pump and the CGM.
We did the first trials with this system in Italy and France in the fall of last year, and we did the first U.S.-based trial from April 19 to 21. It seems promising. We did the study in a local hotel. The trial was set up to give the patient a sense of being out on his own without wires for the connection or syringes. It was an interesting experience for the patient and us as well.
DiabetesCare.net: Are you getting individuals’ blood sugar readings ahead of time and then programming it in? Do the algorithms work for everybody or do they have to be personalized?
Kovatchev: The formulas and the algorithms don’t change but it is initialized by the parameters of the patient, so it is personalized in that sense. When we get the person’s parameters such as the basal rate, weight, etc., we manually put them into the system.
You take the cell phone and on the first screen is the entry for these parameters. The smart phone then starts reading the sensor and pump and adapts every five minutes to the reading. Beyond entering the initial parameters, there is no other personalization or need to manually do anything to the phone.
DiabetesCare.net: During the development phase, what was the most challenging element in creating the artificial pancreas? Was it the algorithm in delivering insulin at the right time?
Kovatchev: Yes, the algorithm was challenging. When we started this project years ago, most people did not think it was possible to develop an artificial pancreas due to the delays of insulin action.
It took a few years and the simulation of huge computers looking at the human metabolic system to get where we are today. I think we are practically there now, given the numerous clinical trials we have done. Here at UVA and at our affiliated centers in Europe we have now had over 60 people using the artificial pancreas.
So, we have had time to improve the control algorithm. The next challenge was to make it portable. Just last summer it was unclear whether a cell phone could not run that type of algorithm because it was too complicated. We happily proved this theory wrong very quickly because it turns out the normal smart phone today can run a control algorithm at ten times the speed needed for real-life situations.
Now the challenge that remains is the robustness of the wireless connections between the cell phone, the pump, and the sensor. We are not there yet.
Just in the news today, JDRF reached an agreement with Dexcom to fund them to create a wireless sensor. This sensor should be available by Christmas. That will cover the robustness of the wireless connection.
DiabetesCare.net: Do you have a dominant algorithm that works in the majority of patients and a couple of back-up algorithms? Can you explain what they each do?
Kovatchev: No, we have three different algorithms and they have all been tested in various patient populations.They differ in the aggressiveness of treatment. The one we are running in our outpatient trial is very conservative. It emphasizes patient safety mostly, protecting against hypoglycemia. It is not designed to deliver intensive insulin treatment.
There is a second algorithm that has the safety component but provides more intensive treatment. That is the control-to-range algorithm. That algorithm still allows the person to decide how much bolus insulin is delivered before they eat. It is an adjunct to normal therapy.
Lastly, there is a third algorithm, which is fully-automated. It does not rely on patient input for pre-meal bolus infusions.
DiabetesCare.net: You had mentioned a super computer in helping to develop the algorithms. Were you working with a team of mathematicians and scientists and inputting it into the super computer to confirm the algorithms were working?
Kovatchev: I have been working for many years with a group in Italy. The head of the research there is Dr. Claudio Cobelli. We have been working with them on modeling of the human metabolic system.
When the artificial pancreas research began, the traditional way of testing controlled algorithms was to do animal trials first then move into human trials. This wasn’t good enough for me because it was such a slow, expensive process. I suggested that we should do computer simulation testing as opposed to animal testing. Afterwards, we would go into human testing.
In medical research there is not much computer simulation, but in engineering it is commonplace. In the airline industry, for example, you have flight simulators. Initially, we didn`t have a simulator of the human metabolic system that would allow us to test or control algorithms, so the first thing we did with the University of Padova was to create a human metabolic simulator. At the end of 2007, I submitted the simulator to the FDA for approval.
In January 2008, we got FDA approval, and in May 2008 we had the first human trials in Europe. It accelerated the process quite a bit and made it more inexpensive.
DiabetesCare.net: Can you explain how the computer and the algorithms work?
Kovatchev: The computer is a fast computer, but not necessarily a super computer. It can do the computations of 300 people of different age groups with type 1 diabetes, so in five minutes, you can run through a simulation of 300 subjects and get results.
If an algorithm is not working, it will crash very quickly. On the other hand, if the algorithm passes computer simulation, it doesn’t necessarily mean it is going to work in real life. It is, however, reasonable to test that algorithm in people.
DiabetesCare.net: How long will the outpatient trial last?
Kovatchev: The trial is planned to go on through the end of 2013, and the results should be available in early 2014. This is when we could possibly take it out of the academic setting, upon FDA approval, and pass it off to device manufacturers.
DiabetesCare.net: If you have a successful trial and you are allowed to take it out of the academic setting, what is next?
Kovatchev: The University of Virginia Licensing and Venture Group is talking to various device manufacturers now about developing the product for commercial use. The device companies have to decide to manufacture it and get FDA approval for distribution. My guess is the artificial pancreas would be offered as an adjunct for insulin pump therapy initially, and not as a replacement. Once sufficient confidence is gained over time with the artificial pancreas, it could be used as a replacement device.
DiabetesCare.net: For the closed loop artificial pancreas to be used as a replacement device to the pump and CGM and take over as a first-line, effective therapy for people with type 1 diabetes, what do you think is the timeframe for that to happen?
Kovatchev: I would say the replacement component is going to be several years away mostly because of the need for acceptance by clinicians and patients. Elements such as education, acceptance, and training need to happen first before it becomes used as a mainstream, best practice in treatment therapy.
The photo of Dr. Kovatchev was taken by Dan Addison, UVA Public Affairs.